Giters

daoyuan98 / Relation-CZSL

Official implementation of the TMM paper "Relation-aware Compositional Zero-shot Learning for Attribute-Object Pair Recognition".

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

The test pair set in evaluation is different from other released code

care77 opened this issue · comments

commented

Hi, @ZiweiXU @daoyuan98 . I find that the test pair set for evaluation (line197-204) is less than other released code (line 225-250).
For example,

  1. MIT-States contains 1262 pairs in train set, 300/300 seen/unseen pairs in val set, 400/400 seen/unseen pairs in test set;
  2. when evaluated on test set, test pair set contains 800 pairs in Relation-CZSL but 1262 + 400 pairs in ExplainableML-czsl.

Is it a mistake here?